Refactoring and enhancements to RGBASM warnings

[Rangi42]

Fixes #1527

• Allow a no- prefix to negate "meta" warnings (-Wno-all, -Wno-extra, -Wno-everything)
• Allow -Wno-error to prevent -Werror from making specific individual or meta warnings into errors

Needs more test cases, but ready to review as-is. (test/asm/warn-numeric-string.asm covers a lot already.)

(And yay, -55 LoC!)

[ISSOtm]

I'm not sure that I like OPT w. In what situations would someone want to disable all warnings, programmatically? (I'm already not super fond of the option in the first place.)

[aaaaaa123456789]

I'm not sure that I like OPT w. In what situations would someone want to disable all warnings, programmatically? (I'm already not super fond of the option in the first place.)

Consistency. The mental model for OPT is significantly simpler when you don't have to guess at which options it supports.
I can definitely see it used for shorter code when you want to temporarily disable a few warnings.

[ISSOtm]

But several options are not supported already: -E, for example.

I don't think that OPT should support all CLI options, either, because some of them don't make sense beyond the initial CLI invocation (why use OPT DDUCK=🦆 when def DUCK equs "🦆" is right there?); this would mean that there will always be some inconsistency, and that thus the line should be drawn somewhere.

[aaaaaa123456789]

Sure, "this option makes no sense" is a reasonable argument. "This option is perfectly functional and meaningful, but I don't like it" is a very dubious one.

[ISSOtm]

OPT DDUCK makes sense (“define a string symbol called DUCK and whose contents are 1), and is perfectly functional since EQUS is. It is purely a “I don't like that it would create two string definition syntaxes”.

The rationale for OPT is not “allow changing all/some CLI flags at runtime”, but “allow persisting some changes to the input's processing“.

OPT b is “well, I like my %0010_1010s, but I'd like to use something more visually clear for this one-off 1bpp graphic”. OPT g is the same.

OPT p made sense before we supported 2+-arg ds, and has been kept only for compat since, despite seeing near-zero use (AFAIA, anyway).

OPT W is “this specific line of code triggers a false positive that I'd like to keep enabled for the rest of the program, and I'd like not to get a warning printed on every single build”.

I could go on, but to me, every single one of the currently-supported OPT flags has a reason to be toggled at runtime, beyond “there's an equivalent CLI switch”. And I don't see OPT w passing that test, hence why I am feeling iffy about its addition.

[aaaaaa123456789]

I have absolutely no issues with OPT DDUCK=1, for the record.

[Rangi42]

My two reasons for including OPT w support were consistency and to allow disabling all warnings programmatically. You're right that we're not totally consistent about "all CLI options are OPT options" (and don't need to be), and OPT Wno-everything will allow disabling everything anyway, so I'm okay with removing OPT w if you still prefer that.

[aaaaaa123456789]

I'd say there's a clear line that you can have: things that can be pushed and popped. PUSHO/POPO are critical to the safe usage of OPT, so you probably don't want to add any flags that wouldn't be easy to push and pop. Things like -E fall in this bin.
On the other hand, w is perfectly pushable and poppable.

[Rangi42]

This is the pseudocode logic we should follow, after #1527 is also implemented:

if -Wno-foobar {
	return DISABLED;
} else if -Werror=foobar {
	return ERROR;
} else if -Wfoobar {
	if -Werror and not -Wno-error=foobar and not covered by -Wno-error=meta {
		return ERROR;
	} else {
		return ENABLED;
	}
} else if covered by -Wno-meta {
	return DISABLED;
} else if covered by -Werror=meta {
	return ERROR;
} else if covered by -Wmeta {
	if -Werror and not -Wno-error=foobar and not covered by -Wno-error=meta {
		return ERROR;
	} else {
		return ENABLED;
	}
} else if disabled by default {
	return DISABLED;
} else if errored by default { // never actually happens but technically possible
	return ERROR;
} else { // enabled by default
	if -Werror and not -Wno-error=foobar and not covered by -Wno-error=meta {
		return ERROR;
	} else {
		return ENABLED;
	}
}

[Rangi42]

So, -w is not meant as an alias for -Wno-everything; it's meant to simply disable all warnings, regardless of -W... or current OPT W... state. I think that's fine as just a CLI flag.

[ISSOtm]

I don't fully grok the implementation, but Sylvie walked me through the logic, and it seems sound. That's enough for me, so, merging.